Mastering (?) Deconvolution

[Jack Harvey]

Yep will keep a lookout for your next image.

Yeah we live on a ranch and have horses, but I don't ride as much as I used to because of a old bull riding wreck 15 years ago.  Don't smoke but do drink good whiskey<G>.

Jack
www.Pegasusastronomy.com

[Niall Saunders]

Hi Dave,

You mentioned 'focus issues' in your last post - are these 'general' issues, or did you just happen to have a 'bad session'?

Obviously, if you can keep 'focus' under control, then Deconvolution becomes less of a requirement (hopefully )

Sent from my iPhone - sitting at the departure gate in Heathrow, already suffering PI withdrawal symptoms!!

[dhalliday]

Jack
Well...your M57 is "OK"...but I can always clean it up a little for you,if you want some help...
That's all pretty impressive !!
I will look around a bit more at your site.
So you obviously have "all the keys" to the Pixinsight kingdom...Hmmnn....
Just toying with us amateurs..?!!


Niall
Have a good trip...or are you just hanging out at the airport trying to stay warm??
FOCUS with CCD camera's is a bit tricky!!
My Vixen has an electric focuser,with a 9Volt "toggle switch" thingy...
I can set up a scrollimg "miniview" on my laptop and download a series of brief exposures.
The MAIN problem is its like minus 20...and has been for some time.This is COLD,even for us...
This leads to
1)grinding gears on the EQ6 mount
2)a bunch of stiff (+++) cables(which can snap connector pins on expensive camera's)
3)frost problems on optics
4)FROZEN gears in the focuser.At minus 20 it does NOT move...just a bad grinding sound...!

I actually keep a hairdryer (abandoned by an old flame  ) up in the obs...
I use this to "preheat" the drawtube...and THEN I can SLOWLY move the darn focus..
So its not exactly elegant.You have so many minutes to get focus "close"...then your hands turn blue,then white...then time is up!
Unlike Jack I am not sipping Jack Daniels while "on line" to some balmy remote site...!!!
So all this ( ) to say that focus WAS off the other night.
I certainly do not expect Decon to make a "silk purse out of a sows ear"...
BUT I can honestly say that with my gear I have NEVER seen an M57 as good as mine,DSLR wise...
So this leads to a certain sense of (false I am sure) pride.
Sensing IMPROVED resolution in my SBIG (no bayer matrix) I am keen to get a BETTER result...
(Human nature,etc)
Decon seems a more powerful tool to improve image quality.
I went out and got myself a honking quad core CPU..
The image (data) is small...I can do 100 iterations of decon in like 2 minutes...
THIS SHOULD WORK !!!!!!!!
So...thanks for listening,I feel better.
have a good flight...
I am still hoping to hear more about;
1)What the heck IS deringing
2)What is a support image
3)what is PSF and why do those two sliders seem to CONVERGE  into a certain shape...
More volunteers please !!

Dave

[dhalliday]

Here is Deconvolution applied to nonstretched data...
(observed under a Lum mask of the background,and STF)
http://www.flickr.com/photos/daveh56/4318965041/sizes/l/

Indeed it DID seem to work better,except I think having NOT applied NR,the background got noisy +++
Hence the masking of the background.
Dave

[Silvercup]

Hi Dave:

I've been playing with one of your m57 images to see what I can do, I hope you don't mind. This is the result:



On the left is the original shot, then the deconvoluted one, the third is the image of m57 from CAHA-RECTA-DSA.

I resampled to 2x and applied a HDR, an aggressive deconvolution to red channel, and a smoother one in the green and blue channels. In some internal parts of the nebula the deconvolution has done its work and recover some structures, but the original image lacks a bit more focus and resolution for deconvolution to do its job completely. Due to the lack of resolution and focus, some artifacts are generated as spotty paterns. It also shows that the deconvolution has recovered the definition of stars, including some that are not appreciated in the original shot, indicating that the deconvolution parameters are correct.

Lets play with an image of M57 form internet, for example this (I hope they don't mind):

http://www.schoolsobservatory.org.uk/obs/eg/m57.shtml

The image has good resolution (of course, it is made with a 2 meters telescope), but the definition is not so good, lets see what an HDR and a precise deconvolution can do (there is some SCNR in the green channel and a bit of curves too):



As you can see the deconvolution has recovered all data from the nebula and is almost a clone of the very detailed version of the CAHA-RECTA-DSA image.

The deconvolution has made without deringing support (the support image is usually a starmask and the weakest parts of the image to be deconvoluted). You can watch the Gibbs effect in the central star.

What is my advice? Try to improve the focus as possible, it would be desirable to increase the resolution, shoot a little more focal lenght, try to make a lot of shots and averaging those with better resolution rejecting those in which the stars are not perfectly round. Process image making previously a HDR and after do deconvolution.

I hope that you find helpful.
Silvercup

[mmirot]

Silvercup,

Do you generally just use a star mask for deringing?

Max

[dhalliday]

Silvercup
That is amazing !!
A few questions..
Which M57 image of mine was that ?
One I posted here,or on my Flickr site..?
No matter,just curious.I love (!!) what it did,esp to my stars.
So you would NOT use the deringing features..?
(Not that I understand them..!)
You use a star and Lum mask?
Can we apply two masks?
You split the RGB image apart and process each differently with respect to Decon only,or differently in other ways..?
Do you focus on the strongest signal (ie red,in this case-I presume?)
This is something I have wondered about....
You certainly gave me some ideas,so thanks !

Dave

[dhalliday]

Holy smokes...!
I went back and took an old DSLR shot of M57...one which I had processed +++
http://www.flickr.com/photos/daveh56/4322815263/
THEN I extracted all 3 channels,and applied deconvolution to the.
I did not use a star mask.
I DID use deringing...
Here is an example of the ? blue,BEFORE,and after;
http://www.flickr.com/photos/daveh56/4323550538/
and after
http://www.flickr.com/photos/daveh56/4323550830/

and on recombo...a better result!!
http://www.flickr.com/photos/daveh56/4322816755/
I am thrilled,and will play around more..!!
This is just a preliminary result...color is all exagerated etc,...

BUT...this is a breakthrough....!!!!!!!!
Prior to using the mono camera I had NEVER even thought of splitting the image data into 3 separate subs...

I am a happy camper!

Dave

[Silvercup]

Max:

As a rule a StarMask is a must because of the Gibbs effect and the posibility of "burning" bigger stars. Depending on the image it can be necessary to protect the faint areas with a lum mask of this areas to avoid artifacts (spotty paterns), so you only need to sum StarMask and Lum Mask with pixelmath to generate the Main Mask. In particular cases with a few iterations and small doconvolution parameters normal deringing is enought.

Dave:

Your image is http://www.flickr.com/photos/daveh56/4299719863/

I use deringing support in your image but no local deringing with a StarMask as it wasn't necesary because the normal deringing did its work. The Lum mask wasn't necessary for the same reason. The spotty patern is due to excesive iterations (I did 30) and of course it's a JPG image.

Yes, I splited RGB channels and did different deconvolution process on each one. You can do deconvolutions on RGB images with similar results. If you don't unchek the Luminance parameter deconvolution is only processed over the Luminance channel of image, but if you uncheck it is done on R , G and B channels separately.

You can try with your image: HDRW on Layer 2 and check To Luminace, apply on RGB image, split RGB channels and do a less agressive deconvolutions with Stdev 1.70 and Shape 0.40, 8 ierations on Red, 2 iterations on Green and Blue, and recombine. You increase definition avoiding artifacts (you can do even a 1 iteration deconvolution to this RGB image with great results). You can do it in L channel too and after do a LRGB combination and use the chrominance noise reduction options with great results.

Why I split channels? Well, do deconvolution on RGB image with the same parameters and 8 iterations (as in Red channel), you increase details but deringing and artifacts too (in this particular case).

I'm glad this can be helpful to everybody.

Best, Silvercup

PD. We were writing at the same time 

[dhalliday]

Silvercup
Thanks...that version of M57 was a good choice...because I had just learned to do DBE properly.!
Feel free to play with ANY image on my site...!
I am learning a lot and very excited by this !!
Again thanks...but I will have a few more questions on Deconvolutions yet !

Dave
PS; Its the first time I hear of ADDING two masks together...
Do you have some pics I/we can see..
PSS This week (?thursday/friday AM I hope to get new data on M57...)